xref: /linux/kernel/seccomp.c (revision 46e6acfe3501fa938af9c5bd730f0020235b08a2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/kernel/seccomp.c
4  *
5  * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
6  *
7  * Copyright (C) 2012 Google, Inc.
8  * Will Drewry <wad@chromium.org>
9  *
10  * This defines a simple but solid secure-computing facility.
11  *
12  * Mode 1 uses a fixed list of allowed system calls.
13  * Mode 2 allows user-defined system call filters in the form
14  *        of Berkeley Packet Filters/Linux Socket Filters.
15  */
16 #define pr_fmt(fmt) "seccomp: " fmt
17 
18 #include <linux/refcount.h>
19 #include <linux/audit.h>
20 #include <linux/compat.h>
21 #include <linux/coredump.h>
22 #include <linux/kmemleak.h>
23 #include <linux/nospec.h>
24 #include <linux/prctl.h>
25 #include <linux/sched.h>
26 #include <linux/sched/task_stack.h>
27 #include <linux/seccomp.h>
28 #include <linux/slab.h>
29 #include <linux/syscalls.h>
30 #include <linux/sysctl.h>
31 
32 /* Not exposed in headers: strictly internal use only. */
33 #define SECCOMP_MODE_DEAD	(SECCOMP_MODE_FILTER + 1)
34 
35 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
36 #include <asm/syscall.h>
37 #endif
38 
39 #ifdef CONFIG_SECCOMP_FILTER
40 #include <linux/file.h>
41 #include <linux/filter.h>
42 #include <linux/pid.h>
43 #include <linux/ptrace.h>
44 #include <linux/capability.h>
45 #include <linux/uaccess.h>
46 #include <linux/anon_inodes.h>
47 #include <linux/lockdep.h>
48 
49 /*
50  * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
51  * wrong direction flag in the ioctl number. This is the broken one,
52  * which the kernel needs to keep supporting until all userspaces stop
53  * using the wrong command number.
54  */
55 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR	SECCOMP_IOR(2, __u64)
56 
57 enum notify_state {
58 	SECCOMP_NOTIFY_INIT,
59 	SECCOMP_NOTIFY_SENT,
60 	SECCOMP_NOTIFY_REPLIED,
61 };
62 
63 struct seccomp_knotif {
64 	/* The struct pid of the task whose filter triggered the notification */
65 	struct task_struct *task;
66 
67 	/* The "cookie" for this request; this is unique for this filter. */
68 	u64 id;
69 
70 	/*
71 	 * The seccomp data. This pointer is valid the entire time this
72 	 * notification is active, since it comes from __seccomp_filter which
73 	 * eclipses the entire lifecycle here.
74 	 */
75 	const struct seccomp_data *data;
76 
77 	/*
78 	 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
79 	 * struct seccomp_knotif is created and starts out in INIT. Once the
80 	 * handler reads the notification off of an FD, it transitions to SENT.
81 	 * If a signal is received the state transitions back to INIT and
82 	 * another message is sent. When the userspace handler replies, state
83 	 * transitions to REPLIED.
84 	 */
85 	enum notify_state state;
86 
87 	/* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
88 	int error;
89 	long val;
90 	u32 flags;
91 
92 	/*
93 	 * Signals when this has changed states, such as the listener
94 	 * dying, a new seccomp addfd message, or changing to REPLIED
95 	 */
96 	struct completion ready;
97 
98 	struct list_head list;
99 
100 	/* outstanding addfd requests */
101 	struct list_head addfd;
102 };
103 
104 /**
105  * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
106  *
107  * @file: A reference to the file to install in the other task
108  * @fd: The fd number to install it at. If the fd number is -1, it means the
109  *      installing process should allocate the fd as normal.
110  * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
111  *         is allowed.
112  * @ioctl_flags: The flags used for the seccomp_addfd ioctl.
113  * @setfd: whether or not SECCOMP_ADDFD_FLAG_SETFD was set during notify_addfd
114  * @ret: The return value of the installing process. It is set to the fd num
115  *       upon success (>= 0).
116  * @completion: Indicates that the installing process has completed fd
117  *              installation, or gone away (either due to successful
118  *              reply, or signal)
119  * @list: list_head for chaining seccomp_kaddfd together.
120  *
121  */
122 struct seccomp_kaddfd {
123 	struct file *file;
124 	int fd;
125 	unsigned int flags;
126 	__u32 ioctl_flags;
127 
128 	union {
129 		bool setfd;
130 		/* To only be set on reply */
131 		int ret;
132 	};
133 	struct completion completion;
134 	struct list_head list;
135 };
136 
137 /**
138  * struct notification - container for seccomp userspace notifications. Since
139  * most seccomp filters will not have notification listeners attached and this
140  * structure is fairly large, we store the notification-specific stuff in a
141  * separate structure.
142  *
143  * @requests: A semaphore that users of this notification can wait on for
144  *            changes. Actual reads and writes are still controlled with
145  *            filter->notify_lock.
146  * @flags: A set of SECCOMP_USER_NOTIF_FD_* flags.
147  * @next_id: The id of the next request.
148  * @notifications: A list of struct seccomp_knotif elements.
149  */
150 
151 struct notification {
152 	atomic_t requests;
153 	u32 flags;
154 	u64 next_id;
155 	struct list_head notifications;
156 };
157 
158 #ifdef SECCOMP_ARCH_NATIVE
159 /**
160  * struct action_cache - per-filter cache of seccomp actions per
161  * arch/syscall pair
162  *
163  * @allow_native: A bitmap where each bit represents whether the
164  *		  filter will always allow the syscall, for the
165  *		  native architecture.
166  * @allow_compat: A bitmap where each bit represents whether the
167  *		  filter will always allow the syscall, for the
168  *		  compat architecture.
169  */
170 struct action_cache {
171 	DECLARE_BITMAP(allow_native, SECCOMP_ARCH_NATIVE_NR);
172 #ifdef SECCOMP_ARCH_COMPAT
173 	DECLARE_BITMAP(allow_compat, SECCOMP_ARCH_COMPAT_NR);
174 #endif
175 };
176 #else
177 struct action_cache { };
178 
179 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
180 					     const struct seccomp_data *sd)
181 {
182 	return false;
183 }
184 
185 static inline void seccomp_cache_prepare(struct seccomp_filter *sfilter)
186 {
187 }
188 #endif /* SECCOMP_ARCH_NATIVE */
189 
190 /**
191  * struct seccomp_filter - container for seccomp BPF programs
192  *
193  * @refs: Reference count to manage the object lifetime.
194  *	  A filter's reference count is incremented for each directly
195  *	  attached task, once for the dependent filter, and if
196  *	  requested for the user notifier. When @refs reaches zero,
197  *	  the filter can be freed.
198  * @users: A filter's @users count is incremented for each directly
199  *         attached task (filter installation, fork(), thread_sync),
200  *	   and once for the dependent filter (tracked in filter->prev).
201  *	   When it reaches zero it indicates that no direct or indirect
202  *	   users of that filter exist. No new tasks can get associated with
203  *	   this filter after reaching 0. The @users count is always smaller
204  *	   or equal to @refs. Hence, reaching 0 for @users does not mean
205  *	   the filter can be freed.
206  * @cache: cache of arch/syscall mappings to actions
207  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
208  * @wait_killable_recv: Put notifying process in killable state once the
209  *			notification is received by the userspace listener.
210  * @prev: points to a previously installed, or inherited, filter
211  * @prog: the BPF program to evaluate
212  * @notif: the struct that holds all notification related information
213  * @notify_lock: A lock for all notification-related accesses.
214  * @wqh: A wait queue for poll if a notifier is in use.
215  *
216  * seccomp_filter objects are organized in a tree linked via the @prev
217  * pointer.  For any task, it appears to be a singly-linked list starting
218  * with current->seccomp.filter, the most recently attached or inherited filter.
219  * However, multiple filters may share a @prev node, by way of fork(), which
220  * results in a unidirectional tree existing in memory.  This is similar to
221  * how namespaces work.
222  *
223  * seccomp_filter objects should never be modified after being attached
224  * to a task_struct (other than @refs).
225  */
226 struct seccomp_filter {
227 	refcount_t refs;
228 	refcount_t users;
229 	bool log;
230 	bool wait_killable_recv;
231 	struct action_cache cache;
232 	struct seccomp_filter *prev;
233 	struct bpf_prog *prog;
234 	struct notification *notif;
235 	struct mutex notify_lock;
236 	wait_queue_head_t wqh;
237 };
238 
239 /* Limit any path through the tree to 256KB worth of instructions. */
240 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
241 
242 /*
243  * Endianness is explicitly ignored and left for BPF program authors to manage
244  * as per the specific architecture.
245  */
246 static void populate_seccomp_data(struct seccomp_data *sd)
247 {
248 	/*
249 	 * Instead of using current_pt_reg(), we're already doing the work
250 	 * to safely fetch "current", so just use "task" everywhere below.
251 	 */
252 	struct task_struct *task = current;
253 	struct pt_regs *regs = task_pt_regs(task);
254 	unsigned long args[6];
255 
256 	sd->nr = syscall_get_nr(task, regs);
257 	sd->arch = syscall_get_arch(task);
258 	syscall_get_arguments(task, regs, args);
259 	sd->args[0] = args[0];
260 	sd->args[1] = args[1];
261 	sd->args[2] = args[2];
262 	sd->args[3] = args[3];
263 	sd->args[4] = args[4];
264 	sd->args[5] = args[5];
265 	sd->instruction_pointer = KSTK_EIP(task);
266 }
267 
268 /**
269  *	seccomp_check_filter - verify seccomp filter code
270  *	@filter: filter to verify
271  *	@flen: length of filter
272  *
273  * Takes a previously checked filter (by bpf_check_classic) and
274  * redirects all filter code that loads struct sk_buff data
275  * and related data through seccomp_bpf_load.  It also
276  * enforces length and alignment checking of those loads.
277  *
278  * Returns 0 if the rule set is legal or -EINVAL if not.
279  */
280 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
281 {
282 	int pc;
283 	for (pc = 0; pc < flen; pc++) {
284 		struct sock_filter *ftest = &filter[pc];
285 		u16 code = ftest->code;
286 		u32 k = ftest->k;
287 
288 		switch (code) {
289 		case BPF_LD | BPF_W | BPF_ABS:
290 			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
291 			/* 32-bit aligned and not out of bounds. */
292 			if (k >= sizeof(struct seccomp_data) || k & 3)
293 				return -EINVAL;
294 			continue;
295 		case BPF_LD | BPF_W | BPF_LEN:
296 			ftest->code = BPF_LD | BPF_IMM;
297 			ftest->k = sizeof(struct seccomp_data);
298 			continue;
299 		case BPF_LDX | BPF_W | BPF_LEN:
300 			ftest->code = BPF_LDX | BPF_IMM;
301 			ftest->k = sizeof(struct seccomp_data);
302 			continue;
303 		/* Explicitly include allowed calls. */
304 		case BPF_RET | BPF_K:
305 		case BPF_RET | BPF_A:
306 		case BPF_ALU | BPF_ADD | BPF_K:
307 		case BPF_ALU | BPF_ADD | BPF_X:
308 		case BPF_ALU | BPF_SUB | BPF_K:
309 		case BPF_ALU | BPF_SUB | BPF_X:
310 		case BPF_ALU | BPF_MUL | BPF_K:
311 		case BPF_ALU | BPF_MUL | BPF_X:
312 		case BPF_ALU | BPF_DIV | BPF_K:
313 		case BPF_ALU | BPF_DIV | BPF_X:
314 		case BPF_ALU | BPF_AND | BPF_K:
315 		case BPF_ALU | BPF_AND | BPF_X:
316 		case BPF_ALU | BPF_OR | BPF_K:
317 		case BPF_ALU | BPF_OR | BPF_X:
318 		case BPF_ALU | BPF_XOR | BPF_K:
319 		case BPF_ALU | BPF_XOR | BPF_X:
320 		case BPF_ALU | BPF_LSH | BPF_K:
321 		case BPF_ALU | BPF_LSH | BPF_X:
322 		case BPF_ALU | BPF_RSH | BPF_K:
323 		case BPF_ALU | BPF_RSH | BPF_X:
324 		case BPF_ALU | BPF_NEG:
325 		case BPF_LD | BPF_IMM:
326 		case BPF_LDX | BPF_IMM:
327 		case BPF_MISC | BPF_TAX:
328 		case BPF_MISC | BPF_TXA:
329 		case BPF_LD | BPF_MEM:
330 		case BPF_LDX | BPF_MEM:
331 		case BPF_ST:
332 		case BPF_STX:
333 		case BPF_JMP | BPF_JA:
334 		case BPF_JMP | BPF_JEQ | BPF_K:
335 		case BPF_JMP | BPF_JEQ | BPF_X:
336 		case BPF_JMP | BPF_JGE | BPF_K:
337 		case BPF_JMP | BPF_JGE | BPF_X:
338 		case BPF_JMP | BPF_JGT | BPF_K:
339 		case BPF_JMP | BPF_JGT | BPF_X:
340 		case BPF_JMP | BPF_JSET | BPF_K:
341 		case BPF_JMP | BPF_JSET | BPF_X:
342 			continue;
343 		default:
344 			return -EINVAL;
345 		}
346 	}
347 	return 0;
348 }
349 
350 #ifdef SECCOMP_ARCH_NATIVE
351 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap,
352 						    size_t bitmap_size,
353 						    int syscall_nr)
354 {
355 	if (unlikely(syscall_nr < 0 || syscall_nr >= bitmap_size))
356 		return false;
357 	syscall_nr = array_index_nospec(syscall_nr, bitmap_size);
358 
359 	return test_bit(syscall_nr, bitmap);
360 }
361 
362 /**
363  * seccomp_cache_check_allow - lookup seccomp cache
364  * @sfilter: The seccomp filter
365  * @sd: The seccomp data to lookup the cache with
366  *
367  * Returns true if the seccomp_data is cached and allowed.
368  */
369 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter,
370 					     const struct seccomp_data *sd)
371 {
372 	int syscall_nr = sd->nr;
373 	const struct action_cache *cache = &sfilter->cache;
374 
375 #ifndef SECCOMP_ARCH_COMPAT
376 	/* A native-only architecture doesn't need to check sd->arch. */
377 	return seccomp_cache_check_allow_bitmap(cache->allow_native,
378 						SECCOMP_ARCH_NATIVE_NR,
379 						syscall_nr);
380 #else
381 	if (likely(sd->arch == SECCOMP_ARCH_NATIVE))
382 		return seccomp_cache_check_allow_bitmap(cache->allow_native,
383 							SECCOMP_ARCH_NATIVE_NR,
384 							syscall_nr);
385 	if (likely(sd->arch == SECCOMP_ARCH_COMPAT))
386 		return seccomp_cache_check_allow_bitmap(cache->allow_compat,
387 							SECCOMP_ARCH_COMPAT_NR,
388 							syscall_nr);
389 #endif /* SECCOMP_ARCH_COMPAT */
390 
391 	WARN_ON_ONCE(true);
392 	return false;
393 }
394 #endif /* SECCOMP_ARCH_NATIVE */
395 
396 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
397 /**
398  * seccomp_run_filters - evaluates all seccomp filters against @sd
399  * @sd: optional seccomp data to be passed to filters
400  * @match: stores struct seccomp_filter that resulted in the return value,
401  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
402  *         be unchanged.
403  *
404  * Returns valid seccomp BPF response codes.
405  */
406 static u32 seccomp_run_filters(const struct seccomp_data *sd,
407 			       struct seccomp_filter **match)
408 {
409 	u32 ret = SECCOMP_RET_ALLOW;
410 	/* Make sure cross-thread synced filter points somewhere sane. */
411 	struct seccomp_filter *f =
412 			READ_ONCE(current->seccomp.filter);
413 
414 	/* Ensure unexpected behavior doesn't result in failing open. */
415 	if (WARN_ON(f == NULL))
416 		return SECCOMP_RET_KILL_PROCESS;
417 
418 	if (seccomp_cache_check_allow(f, sd))
419 		return SECCOMP_RET_ALLOW;
420 
421 	/*
422 	 * All filters in the list are evaluated and the lowest BPF return
423 	 * value always takes priority (ignoring the DATA).
424 	 */
425 	for (; f; f = f->prev) {
426 		u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
427 
428 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
429 			ret = cur_ret;
430 			*match = f;
431 		}
432 	}
433 	return ret;
434 }
435 #endif /* CONFIG_SECCOMP_FILTER */
436 
437 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
438 {
439 	assert_spin_locked(&current->sighand->siglock);
440 
441 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
442 		return false;
443 
444 	return true;
445 }
446 
447 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
448 
449 static inline void seccomp_assign_mode(struct task_struct *task,
450 				       unsigned long seccomp_mode,
451 				       unsigned long flags)
452 {
453 	assert_spin_locked(&task->sighand->siglock);
454 
455 	task->seccomp.mode = seccomp_mode;
456 	/*
457 	 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and
458 	 * filter) is set.
459 	 */
460 	smp_mb__before_atomic();
461 	/* Assume default seccomp processes want spec flaw mitigation. */
462 	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
463 		arch_seccomp_spec_mitigate(task);
464 	set_task_syscall_work(task, SECCOMP);
465 }
466 
467 #ifdef CONFIG_SECCOMP_FILTER
468 /* Returns 1 if the parent is an ancestor of the child. */
469 static int is_ancestor(struct seccomp_filter *parent,
470 		       struct seccomp_filter *child)
471 {
472 	/* NULL is the root ancestor. */
473 	if (parent == NULL)
474 		return 1;
475 	for (; child; child = child->prev)
476 		if (child == parent)
477 			return 1;
478 	return 0;
479 }
480 
481 /**
482  * seccomp_can_sync_threads: checks if all threads can be synchronized
483  *
484  * Expects sighand and cred_guard_mutex locks to be held.
485  *
486  * Returns 0 on success, -ve on error, or the pid of a thread which was
487  * either not in the correct seccomp mode or did not have an ancestral
488  * seccomp filter.
489  */
490 static inline pid_t seccomp_can_sync_threads(void)
491 {
492 	struct task_struct *thread, *caller;
493 
494 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
495 	assert_spin_locked(&current->sighand->siglock);
496 
497 	/* Validate all threads being eligible for synchronization. */
498 	caller = current;
499 	for_each_thread(caller, thread) {
500 		pid_t failed;
501 
502 		/* Skip current, since it is initiating the sync. */
503 		if (thread == caller)
504 			continue;
505 		/* Skip exited threads. */
506 		if (thread->flags & PF_EXITING)
507 			continue;
508 
509 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
510 		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
511 		     is_ancestor(thread->seccomp.filter,
512 				 caller->seccomp.filter)))
513 			continue;
514 
515 		/* Return the first thread that cannot be synchronized. */
516 		failed = task_pid_vnr(thread);
517 		/* If the pid cannot be resolved, then return -ESRCH */
518 		if (WARN_ON(failed == 0))
519 			failed = -ESRCH;
520 		return failed;
521 	}
522 
523 	return 0;
524 }
525 
526 static inline void seccomp_filter_free(struct seccomp_filter *filter)
527 {
528 	if (filter) {
529 		bpf_prog_destroy(filter->prog);
530 		kfree(filter);
531 	}
532 }
533 
534 static void __seccomp_filter_orphan(struct seccomp_filter *orig)
535 {
536 	while (orig && refcount_dec_and_test(&orig->users)) {
537 		if (waitqueue_active(&orig->wqh))
538 			wake_up_poll(&orig->wqh, EPOLLHUP);
539 		orig = orig->prev;
540 	}
541 }
542 
543 static void __put_seccomp_filter(struct seccomp_filter *orig)
544 {
545 	/* Clean up single-reference branches iteratively. */
546 	while (orig && refcount_dec_and_test(&orig->refs)) {
547 		struct seccomp_filter *freeme = orig;
548 		orig = orig->prev;
549 		seccomp_filter_free(freeme);
550 	}
551 }
552 
553 static void __seccomp_filter_release(struct seccomp_filter *orig)
554 {
555 	/* Notify about any unused filters in the task's former filter tree. */
556 	__seccomp_filter_orphan(orig);
557 	/* Finally drop all references to the task's former tree. */
558 	__put_seccomp_filter(orig);
559 }
560 
561 /**
562  * seccomp_filter_release - Detach the task from its filter tree,
563  *			    drop its reference count, and notify
564  *			    about unused filters
565  *
566  * @tsk: task the filter should be released from.
567  *
568  * This function should only be called when the task is exiting as
569  * it detaches it from its filter tree. PF_EXITING has to be set
570  * for the task.
571  */
572 void seccomp_filter_release(struct task_struct *tsk)
573 {
574 	struct seccomp_filter *orig;
575 
576 	if (WARN_ON((tsk->flags & PF_EXITING) == 0))
577 		return;
578 
579 	spin_lock_irq(&tsk->sighand->siglock);
580 	orig = tsk->seccomp.filter;
581 	/* Detach task from its filter tree. */
582 	tsk->seccomp.filter = NULL;
583 	spin_unlock_irq(&tsk->sighand->siglock);
584 	__seccomp_filter_release(orig);
585 }
586 
587 /**
588  * seccomp_sync_threads: sets all threads to use current's filter
589  *
590  * @flags: SECCOMP_FILTER_FLAG_* flags to set during sync.
591  *
592  * Expects sighand and cred_guard_mutex locks to be held, and for
593  * seccomp_can_sync_threads() to have returned success already
594  * without dropping the locks.
595  *
596  */
597 static inline void seccomp_sync_threads(unsigned long flags)
598 {
599 	struct task_struct *thread, *caller;
600 
601 	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
602 	assert_spin_locked(&current->sighand->siglock);
603 
604 	/* Synchronize all threads. */
605 	caller = current;
606 	for_each_thread(caller, thread) {
607 		/* Skip current, since it needs no changes. */
608 		if (thread == caller)
609 			continue;
610 
611 		/*
612 		 * Skip exited threads. seccomp_filter_release could have
613 		 * been already called for this task.
614 		 */
615 		if (thread->flags & PF_EXITING)
616 			continue;
617 
618 		/* Get a task reference for the new leaf node. */
619 		get_seccomp_filter(caller);
620 
621 		/*
622 		 * Drop the task reference to the shared ancestor since
623 		 * current's path will hold a reference.  (This also
624 		 * allows a put before the assignment.)
625 		 */
626 		__seccomp_filter_release(thread->seccomp.filter);
627 
628 		/* Make our new filter tree visible. */
629 		smp_store_release(&thread->seccomp.filter,
630 				  caller->seccomp.filter);
631 		atomic_set(&thread->seccomp.filter_count,
632 			   atomic_read(&caller->seccomp.filter_count));
633 
634 		/*
635 		 * Don't let an unprivileged task work around
636 		 * the no_new_privs restriction by creating
637 		 * a thread that sets it up, enters seccomp,
638 		 * then dies.
639 		 */
640 		if (task_no_new_privs(caller))
641 			task_set_no_new_privs(thread);
642 
643 		/*
644 		 * Opt the other thread into seccomp if needed.
645 		 * As threads are considered to be trust-realm
646 		 * equivalent (see ptrace_may_access), it is safe to
647 		 * allow one thread to transition the other.
648 		 */
649 		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
650 			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
651 					    flags);
652 	}
653 }
654 
655 /**
656  * seccomp_prepare_filter: Prepares a seccomp filter for use.
657  * @fprog: BPF program to install
658  *
659  * Returns filter on success or an ERR_PTR on failure.
660  */
661 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
662 {
663 	struct seccomp_filter *sfilter;
664 	int ret;
665 	const bool save_orig =
666 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE)
667 		true;
668 #else
669 		false;
670 #endif
671 
672 	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
673 		return ERR_PTR(-EINVAL);
674 
675 	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
676 
677 	/*
678 	 * Installing a seccomp filter requires that the task has
679 	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
680 	 * This avoids scenarios where unprivileged tasks can affect the
681 	 * behavior of privileged children.
682 	 */
683 	if (!task_no_new_privs(current) &&
684 			!ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
685 		return ERR_PTR(-EACCES);
686 
687 	/* Allocate a new seccomp_filter */
688 	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
689 	if (!sfilter)
690 		return ERR_PTR(-ENOMEM);
691 
692 	mutex_init(&sfilter->notify_lock);
693 	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
694 					seccomp_check_filter, save_orig);
695 	if (ret < 0) {
696 		kfree(sfilter);
697 		return ERR_PTR(ret);
698 	}
699 
700 	refcount_set(&sfilter->refs, 1);
701 	refcount_set(&sfilter->users, 1);
702 	init_waitqueue_head(&sfilter->wqh);
703 
704 	return sfilter;
705 }
706 
707 /**
708  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
709  * @user_filter: pointer to the user data containing a sock_fprog.
710  *
711  * Returns 0 on success and non-zero otherwise.
712  */
713 static struct seccomp_filter *
714 seccomp_prepare_user_filter(const char __user *user_filter)
715 {
716 	struct sock_fprog fprog;
717 	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
718 
719 #ifdef CONFIG_COMPAT
720 	if (in_compat_syscall()) {
721 		struct compat_sock_fprog fprog32;
722 		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
723 			goto out;
724 		fprog.len = fprog32.len;
725 		fprog.filter = compat_ptr(fprog32.filter);
726 	} else /* falls through to the if below. */
727 #endif
728 	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
729 		goto out;
730 	filter = seccomp_prepare_filter(&fprog);
731 out:
732 	return filter;
733 }
734 
735 #ifdef SECCOMP_ARCH_NATIVE
736 /**
737  * seccomp_is_const_allow - check if filter is constant allow with given data
738  * @fprog: The BPF programs
739  * @sd: The seccomp data to check against, only syscall number and arch
740  *      number are considered constant.
741  */
742 static bool seccomp_is_const_allow(struct sock_fprog_kern *fprog,
743 				   struct seccomp_data *sd)
744 {
745 	unsigned int reg_value = 0;
746 	unsigned int pc;
747 	bool op_res;
748 
749 	if (WARN_ON_ONCE(!fprog))
750 		return false;
751 
752 	for (pc = 0; pc < fprog->len; pc++) {
753 		struct sock_filter *insn = &fprog->filter[pc];
754 		u16 code = insn->code;
755 		u32 k = insn->k;
756 
757 		switch (code) {
758 		case BPF_LD | BPF_W | BPF_ABS:
759 			switch (k) {
760 			case offsetof(struct seccomp_data, nr):
761 				reg_value = sd->nr;
762 				break;
763 			case offsetof(struct seccomp_data, arch):
764 				reg_value = sd->arch;
765 				break;
766 			default:
767 				/* can't optimize (non-constant value load) */
768 				return false;
769 			}
770 			break;
771 		case BPF_RET | BPF_K:
772 			/* reached return with constant values only, check allow */
773 			return k == SECCOMP_RET_ALLOW;
774 		case BPF_JMP | BPF_JA:
775 			pc += insn->k;
776 			break;
777 		case BPF_JMP | BPF_JEQ | BPF_K:
778 		case BPF_JMP | BPF_JGE | BPF_K:
779 		case BPF_JMP | BPF_JGT | BPF_K:
780 		case BPF_JMP | BPF_JSET | BPF_K:
781 			switch (BPF_OP(code)) {
782 			case BPF_JEQ:
783 				op_res = reg_value == k;
784 				break;
785 			case BPF_JGE:
786 				op_res = reg_value >= k;
787 				break;
788 			case BPF_JGT:
789 				op_res = reg_value > k;
790 				break;
791 			case BPF_JSET:
792 				op_res = !!(reg_value & k);
793 				break;
794 			default:
795 				/* can't optimize (unknown jump) */
796 				return false;
797 			}
798 
799 			pc += op_res ? insn->jt : insn->jf;
800 			break;
801 		case BPF_ALU | BPF_AND | BPF_K:
802 			reg_value &= k;
803 			break;
804 		default:
805 			/* can't optimize (unknown insn) */
806 			return false;
807 		}
808 	}
809 
810 	/* ran off the end of the filter?! */
811 	WARN_ON(1);
812 	return false;
813 }
814 
815 static void seccomp_cache_prepare_bitmap(struct seccomp_filter *sfilter,
816 					 void *bitmap, const void *bitmap_prev,
817 					 size_t bitmap_size, int arch)
818 {
819 	struct sock_fprog_kern *fprog = sfilter->prog->orig_prog;
820 	struct seccomp_data sd;
821 	int nr;
822 
823 	if (bitmap_prev) {
824 		/* The new filter must be as restrictive as the last. */
825 		bitmap_copy(bitmap, bitmap_prev, bitmap_size);
826 	} else {
827 		/* Before any filters, all syscalls are always allowed. */
828 		bitmap_fill(bitmap, bitmap_size);
829 	}
830 
831 	for (nr = 0; nr < bitmap_size; nr++) {
832 		/* No bitmap change: not a cacheable action. */
833 		if (!test_bit(nr, bitmap))
834 			continue;
835 
836 		sd.nr = nr;
837 		sd.arch = arch;
838 
839 		/* No bitmap change: continue to always allow. */
840 		if (seccomp_is_const_allow(fprog, &sd))
841 			continue;
842 
843 		/*
844 		 * Not a cacheable action: always run filters.
845 		 * atomic clear_bit() not needed, filter not visible yet.
846 		 */
847 		__clear_bit(nr, bitmap);
848 	}
849 }
850 
851 /**
852  * seccomp_cache_prepare - emulate the filter to find cacheable syscalls
853  * @sfilter: The seccomp filter
854  *
855  * Returns 0 if successful or -errno if error occurred.
856  */
857 static void seccomp_cache_prepare(struct seccomp_filter *sfilter)
858 {
859 	struct action_cache *cache = &sfilter->cache;
860 	const struct action_cache *cache_prev =
861 		sfilter->prev ? &sfilter->prev->cache : NULL;
862 
863 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_native,
864 				     cache_prev ? cache_prev->allow_native : NULL,
865 				     SECCOMP_ARCH_NATIVE_NR,
866 				     SECCOMP_ARCH_NATIVE);
867 
868 #ifdef SECCOMP_ARCH_COMPAT
869 	seccomp_cache_prepare_bitmap(sfilter, cache->allow_compat,
870 				     cache_prev ? cache_prev->allow_compat : NULL,
871 				     SECCOMP_ARCH_COMPAT_NR,
872 				     SECCOMP_ARCH_COMPAT);
873 #endif /* SECCOMP_ARCH_COMPAT */
874 }
875 #endif /* SECCOMP_ARCH_NATIVE */
876 
877 /**
878  * seccomp_attach_filter: validate and attach filter
879  * @flags:  flags to change filter behavior
880  * @filter: seccomp filter to add to the current process
881  *
882  * Caller must be holding current->sighand->siglock lock.
883  *
884  * Returns 0 on success, -ve on error, or
885  *   - in TSYNC mode: the pid of a thread which was either not in the correct
886  *     seccomp mode or did not have an ancestral seccomp filter
887  *   - in NEW_LISTENER mode: the fd of the new listener
888  */
889 static long seccomp_attach_filter(unsigned int flags,
890 				  struct seccomp_filter *filter)
891 {
892 	unsigned long total_insns;
893 	struct seccomp_filter *walker;
894 
895 	assert_spin_locked(&current->sighand->siglock);
896 
897 	/* Validate resulting filter length. */
898 	total_insns = filter->prog->len;
899 	for (walker = current->seccomp.filter; walker; walker = walker->prev)
900 		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
901 	if (total_insns > MAX_INSNS_PER_PATH)
902 		return -ENOMEM;
903 
904 	/* If thread sync has been requested, check that it is possible. */
905 	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
906 		int ret;
907 
908 		ret = seccomp_can_sync_threads();
909 		if (ret) {
910 			if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
911 				return -ESRCH;
912 			else
913 				return ret;
914 		}
915 	}
916 
917 	/* Set log flag, if present. */
918 	if (flags & SECCOMP_FILTER_FLAG_LOG)
919 		filter->log = true;
920 
921 	/* Set wait killable flag, if present. */
922 	if (flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV)
923 		filter->wait_killable_recv = true;
924 
925 	/*
926 	 * If there is an existing filter, make it the prev and don't drop its
927 	 * task reference.
928 	 */
929 	filter->prev = current->seccomp.filter;
930 	seccomp_cache_prepare(filter);
931 	current->seccomp.filter = filter;
932 	atomic_inc(&current->seccomp.filter_count);
933 
934 	/* Now that the new filter is in place, synchronize to all threads. */
935 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
936 		seccomp_sync_threads(flags);
937 
938 	return 0;
939 }
940 
941 static void __get_seccomp_filter(struct seccomp_filter *filter)
942 {
943 	refcount_inc(&filter->refs);
944 }
945 
946 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
947 void get_seccomp_filter(struct task_struct *tsk)
948 {
949 	struct seccomp_filter *orig = tsk->seccomp.filter;
950 	if (!orig)
951 		return;
952 	__get_seccomp_filter(orig);
953 	refcount_inc(&orig->users);
954 }
955 
956 #endif	/* CONFIG_SECCOMP_FILTER */
957 
958 /* For use with seccomp_actions_logged */
959 #define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
960 #define SECCOMP_LOG_KILL_THREAD		(1 << 1)
961 #define SECCOMP_LOG_TRAP		(1 << 2)
962 #define SECCOMP_LOG_ERRNO		(1 << 3)
963 #define SECCOMP_LOG_TRACE		(1 << 4)
964 #define SECCOMP_LOG_LOG			(1 << 5)
965 #define SECCOMP_LOG_ALLOW		(1 << 6)
966 #define SECCOMP_LOG_USER_NOTIF		(1 << 7)
967 
968 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
969 				    SECCOMP_LOG_KILL_THREAD  |
970 				    SECCOMP_LOG_TRAP  |
971 				    SECCOMP_LOG_ERRNO |
972 				    SECCOMP_LOG_USER_NOTIF |
973 				    SECCOMP_LOG_TRACE |
974 				    SECCOMP_LOG_LOG;
975 
976 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
977 			       bool requested)
978 {
979 	bool log = false;
980 
981 	switch (action) {
982 	case SECCOMP_RET_ALLOW:
983 		break;
984 	case SECCOMP_RET_TRAP:
985 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
986 		break;
987 	case SECCOMP_RET_ERRNO:
988 		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
989 		break;
990 	case SECCOMP_RET_TRACE:
991 		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
992 		break;
993 	case SECCOMP_RET_USER_NOTIF:
994 		log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
995 		break;
996 	case SECCOMP_RET_LOG:
997 		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
998 		break;
999 	case SECCOMP_RET_KILL_THREAD:
1000 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
1001 		break;
1002 	case SECCOMP_RET_KILL_PROCESS:
1003 	default:
1004 		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
1005 	}
1006 
1007 	/*
1008 	 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
1009 	 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
1010 	 * any action from being logged by removing the action name from the
1011 	 * seccomp_actions_logged sysctl.
1012 	 */
1013 	if (!log)
1014 		return;
1015 
1016 	audit_seccomp(syscall, signr, action);
1017 }
1018 
1019 /*
1020  * Secure computing mode 1 allows only read/write/exit/sigreturn.
1021  * To be fully secure this must be combined with rlimit
1022  * to limit the stack allocations too.
1023  */
1024 static const int mode1_syscalls[] = {
1025 	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
1026 	-1, /* negative terminated */
1027 };
1028 
1029 static void __secure_computing_strict(int this_syscall)
1030 {
1031 	const int *allowed_syscalls = mode1_syscalls;
1032 #ifdef CONFIG_COMPAT
1033 	if (in_compat_syscall())
1034 		allowed_syscalls = get_compat_mode1_syscalls();
1035 #endif
1036 	do {
1037 		if (*allowed_syscalls == this_syscall)
1038 			return;
1039 	} while (*++allowed_syscalls != -1);
1040 
1041 #ifdef SECCOMP_DEBUG
1042 	dump_stack();
1043 #endif
1044 	current->seccomp.mode = SECCOMP_MODE_DEAD;
1045 	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
1046 	do_exit(SIGKILL);
1047 }
1048 
1049 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
1050 void secure_computing_strict(int this_syscall)
1051 {
1052 	int mode = current->seccomp.mode;
1053 
1054 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1055 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1056 		return;
1057 
1058 	if (mode == SECCOMP_MODE_DISABLED)
1059 		return;
1060 	else if (mode == SECCOMP_MODE_STRICT)
1061 		__secure_computing_strict(this_syscall);
1062 	else
1063 		BUG();
1064 }
1065 #else
1066 
1067 #ifdef CONFIG_SECCOMP_FILTER
1068 static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
1069 {
1070 	/*
1071 	 * Note: overflow is ok here, the id just needs to be unique per
1072 	 * filter.
1073 	 */
1074 	lockdep_assert_held(&filter->notify_lock);
1075 	return filter->notif->next_id++;
1076 }
1077 
1078 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd, struct seccomp_knotif *n)
1079 {
1080 	int fd;
1081 
1082 	/*
1083 	 * Remove the notification, and reset the list pointers, indicating
1084 	 * that it has been handled.
1085 	 */
1086 	list_del_init(&addfd->list);
1087 	if (!addfd->setfd)
1088 		fd = receive_fd(addfd->file, NULL, addfd->flags);
1089 	else
1090 		fd = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
1091 	addfd->ret = fd;
1092 
1093 	if (addfd->ioctl_flags & SECCOMP_ADDFD_FLAG_SEND) {
1094 		/* If we fail reset and return an error to the notifier */
1095 		if (fd < 0) {
1096 			n->state = SECCOMP_NOTIFY_SENT;
1097 		} else {
1098 			/* Return the FD we just added */
1099 			n->flags = 0;
1100 			n->error = 0;
1101 			n->val = fd;
1102 		}
1103 	}
1104 
1105 	/*
1106 	 * Mark the notification as completed. From this point, addfd mem
1107 	 * might be invalidated and we can't safely read it anymore.
1108 	 */
1109 	complete(&addfd->completion);
1110 }
1111 
1112 static bool should_sleep_killable(struct seccomp_filter *match,
1113 				  struct seccomp_knotif *n)
1114 {
1115 	return match->wait_killable_recv && n->state == SECCOMP_NOTIFY_SENT;
1116 }
1117 
1118 static int seccomp_do_user_notification(int this_syscall,
1119 					struct seccomp_filter *match,
1120 					const struct seccomp_data *sd)
1121 {
1122 	int err;
1123 	u32 flags = 0;
1124 	long ret = 0;
1125 	struct seccomp_knotif n = {};
1126 	struct seccomp_kaddfd *addfd, *tmp;
1127 
1128 	mutex_lock(&match->notify_lock);
1129 	err = -ENOSYS;
1130 	if (!match->notif)
1131 		goto out;
1132 
1133 	n.task = current;
1134 	n.state = SECCOMP_NOTIFY_INIT;
1135 	n.data = sd;
1136 	n.id = seccomp_next_notify_id(match);
1137 	init_completion(&n.ready);
1138 	list_add_tail(&n.list, &match->notif->notifications);
1139 	INIT_LIST_HEAD(&n.addfd);
1140 
1141 	atomic_inc(&match->notif->requests);
1142 	if (match->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1143 		wake_up_poll_on_current_cpu(&match->wqh, EPOLLIN | EPOLLRDNORM);
1144 	else
1145 		wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
1146 
1147 	/*
1148 	 * This is where we wait for a reply from userspace.
1149 	 */
1150 	do {
1151 		bool wait_killable = should_sleep_killable(match, &n);
1152 
1153 		mutex_unlock(&match->notify_lock);
1154 		if (wait_killable)
1155 			err = wait_for_completion_killable(&n.ready);
1156 		else
1157 			err = wait_for_completion_interruptible(&n.ready);
1158 		mutex_lock(&match->notify_lock);
1159 
1160 		if (err != 0) {
1161 			/*
1162 			 * Check to see if the notifcation got picked up and
1163 			 * whether we should switch to wait killable.
1164 			 */
1165 			if (!wait_killable && should_sleep_killable(match, &n))
1166 				continue;
1167 
1168 			goto interrupted;
1169 		}
1170 
1171 		addfd = list_first_entry_or_null(&n.addfd,
1172 						 struct seccomp_kaddfd, list);
1173 		/* Check if we were woken up by a addfd message */
1174 		if (addfd)
1175 			seccomp_handle_addfd(addfd, &n);
1176 
1177 	}  while (n.state != SECCOMP_NOTIFY_REPLIED);
1178 
1179 	ret = n.val;
1180 	err = n.error;
1181 	flags = n.flags;
1182 
1183 interrupted:
1184 	/* If there were any pending addfd calls, clear them out */
1185 	list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
1186 		/* The process went away before we got a chance to handle it */
1187 		addfd->ret = -ESRCH;
1188 		list_del_init(&addfd->list);
1189 		complete(&addfd->completion);
1190 	}
1191 
1192 	/*
1193 	 * Note that it's possible the listener died in between the time when
1194 	 * we were notified of a response (or a signal) and when we were able to
1195 	 * re-acquire the lock, so only delete from the list if the
1196 	 * notification actually exists.
1197 	 *
1198 	 * Also note that this test is only valid because there's no way to
1199 	 * *reattach* to a notifier right now. If one is added, we'll need to
1200 	 * keep track of the notif itself and make sure they match here.
1201 	 */
1202 	if (match->notif)
1203 		list_del(&n.list);
1204 out:
1205 	mutex_unlock(&match->notify_lock);
1206 
1207 	/* Userspace requests to continue the syscall. */
1208 	if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1209 		return 0;
1210 
1211 	syscall_set_return_value(current, current_pt_regs(),
1212 				 err, ret);
1213 	return -1;
1214 }
1215 
1216 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1217 			    const bool recheck_after_trace)
1218 {
1219 	u32 filter_ret, action;
1220 	struct seccomp_filter *match = NULL;
1221 	int data;
1222 	struct seccomp_data sd_local;
1223 
1224 	/*
1225 	 * Make sure that any changes to mode from another thread have
1226 	 * been seen after SYSCALL_WORK_SECCOMP was seen.
1227 	 */
1228 	smp_rmb();
1229 
1230 	if (!sd) {
1231 		populate_seccomp_data(&sd_local);
1232 		sd = &sd_local;
1233 	}
1234 
1235 	filter_ret = seccomp_run_filters(sd, &match);
1236 	data = filter_ret & SECCOMP_RET_DATA;
1237 	action = filter_ret & SECCOMP_RET_ACTION_FULL;
1238 
1239 	switch (action) {
1240 	case SECCOMP_RET_ERRNO:
1241 		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
1242 		if (data > MAX_ERRNO)
1243 			data = MAX_ERRNO;
1244 		syscall_set_return_value(current, current_pt_regs(),
1245 					 -data, 0);
1246 		goto skip;
1247 
1248 	case SECCOMP_RET_TRAP:
1249 		/* Show the handler the original registers. */
1250 		syscall_rollback(current, current_pt_regs());
1251 		/* Let the filter pass back 16 bits of data. */
1252 		force_sig_seccomp(this_syscall, data, false);
1253 		goto skip;
1254 
1255 	case SECCOMP_RET_TRACE:
1256 		/* We've been put in this state by the ptracer already. */
1257 		if (recheck_after_trace)
1258 			return 0;
1259 
1260 		/* ENOSYS these calls if there is no tracer attached. */
1261 		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
1262 			syscall_set_return_value(current,
1263 						 current_pt_regs(),
1264 						 -ENOSYS, 0);
1265 			goto skip;
1266 		}
1267 
1268 		/* Allow the BPF to provide the event message */
1269 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
1270 		/*
1271 		 * The delivery of a fatal signal during event
1272 		 * notification may silently skip tracer notification,
1273 		 * which could leave us with a potentially unmodified
1274 		 * syscall that the tracer would have liked to have
1275 		 * changed. Since the process is about to die, we just
1276 		 * force the syscall to be skipped and let the signal
1277 		 * kill the process and correctly handle any tracer exit
1278 		 * notifications.
1279 		 */
1280 		if (fatal_signal_pending(current))
1281 			goto skip;
1282 		/* Check if the tracer forced the syscall to be skipped. */
1283 		this_syscall = syscall_get_nr(current, current_pt_regs());
1284 		if (this_syscall < 0)
1285 			goto skip;
1286 
1287 		/*
1288 		 * Recheck the syscall, since it may have changed. This
1289 		 * intentionally uses a NULL struct seccomp_data to force
1290 		 * a reload of all registers. This does not goto skip since
1291 		 * a skip would have already been reported.
1292 		 */
1293 		if (__seccomp_filter(this_syscall, NULL, true))
1294 			return -1;
1295 
1296 		return 0;
1297 
1298 	case SECCOMP_RET_USER_NOTIF:
1299 		if (seccomp_do_user_notification(this_syscall, match, sd))
1300 			goto skip;
1301 
1302 		return 0;
1303 
1304 	case SECCOMP_RET_LOG:
1305 		seccomp_log(this_syscall, 0, action, true);
1306 		return 0;
1307 
1308 	case SECCOMP_RET_ALLOW:
1309 		/*
1310 		 * Note that the "match" filter will always be NULL for
1311 		 * this action since SECCOMP_RET_ALLOW is the starting
1312 		 * state in seccomp_run_filters().
1313 		 */
1314 		return 0;
1315 
1316 	case SECCOMP_RET_KILL_THREAD:
1317 	case SECCOMP_RET_KILL_PROCESS:
1318 	default:
1319 		current->seccomp.mode = SECCOMP_MODE_DEAD;
1320 		seccomp_log(this_syscall, SIGSYS, action, true);
1321 		/* Dump core only if this is the last remaining thread. */
1322 		if (action != SECCOMP_RET_KILL_THREAD ||
1323 		    (atomic_read(&current->signal->live) == 1)) {
1324 			/* Show the original registers in the dump. */
1325 			syscall_rollback(current, current_pt_regs());
1326 			/* Trigger a coredump with SIGSYS */
1327 			force_sig_seccomp(this_syscall, data, true);
1328 		} else {
1329 			do_exit(SIGSYS);
1330 		}
1331 		return -1; /* skip the syscall go directly to signal handling */
1332 	}
1333 
1334 	unreachable();
1335 
1336 skip:
1337 	seccomp_log(this_syscall, 0, action, match ? match->log : false);
1338 	return -1;
1339 }
1340 #else
1341 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1342 			    const bool recheck_after_trace)
1343 {
1344 	BUG();
1345 
1346 	return -1;
1347 }
1348 #endif
1349 
1350 int __secure_computing(const struct seccomp_data *sd)
1351 {
1352 	int mode = current->seccomp.mode;
1353 	int this_syscall;
1354 
1355 	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1356 	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1357 		return 0;
1358 
1359 	this_syscall = sd ? sd->nr :
1360 		syscall_get_nr(current, current_pt_regs());
1361 
1362 	switch (mode) {
1363 	case SECCOMP_MODE_STRICT:
1364 		__secure_computing_strict(this_syscall);  /* may call do_exit */
1365 		return 0;
1366 	case SECCOMP_MODE_FILTER:
1367 		return __seccomp_filter(this_syscall, sd, false);
1368 	/* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */
1369 	case SECCOMP_MODE_DEAD:
1370 		WARN_ON_ONCE(1);
1371 		do_exit(SIGKILL);
1372 		return -1;
1373 	default:
1374 		BUG();
1375 	}
1376 }
1377 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1378 
1379 long prctl_get_seccomp(void)
1380 {
1381 	return current->seccomp.mode;
1382 }
1383 
1384 /**
1385  * seccomp_set_mode_strict: internal function for setting strict seccomp
1386  *
1387  * Once current->seccomp.mode is non-zero, it may not be changed.
1388  *
1389  * Returns 0 on success or -EINVAL on failure.
1390  */
1391 static long seccomp_set_mode_strict(void)
1392 {
1393 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1394 	long ret = -EINVAL;
1395 
1396 	spin_lock_irq(&current->sighand->siglock);
1397 
1398 	if (!seccomp_may_assign_mode(seccomp_mode))
1399 		goto out;
1400 
1401 #ifdef TIF_NOTSC
1402 	disable_TSC();
1403 #endif
1404 	seccomp_assign_mode(current, seccomp_mode, 0);
1405 	ret = 0;
1406 
1407 out:
1408 	spin_unlock_irq(&current->sighand->siglock);
1409 
1410 	return ret;
1411 }
1412 
1413 #ifdef CONFIG_SECCOMP_FILTER
1414 static void seccomp_notify_free(struct seccomp_filter *filter)
1415 {
1416 	kfree(filter->notif);
1417 	filter->notif = NULL;
1418 }
1419 
1420 static void seccomp_notify_detach(struct seccomp_filter *filter)
1421 {
1422 	struct seccomp_knotif *knotif;
1423 
1424 	if (!filter)
1425 		return;
1426 
1427 	mutex_lock(&filter->notify_lock);
1428 
1429 	/*
1430 	 * If this file is being closed because e.g. the task who owned it
1431 	 * died, let's wake everyone up who was waiting on us.
1432 	 */
1433 	list_for_each_entry(knotif, &filter->notif->notifications, list) {
1434 		if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1435 			continue;
1436 
1437 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1438 		knotif->error = -ENOSYS;
1439 		knotif->val = 0;
1440 
1441 		/*
1442 		 * We do not need to wake up any pending addfd messages, as
1443 		 * the notifier will do that for us, as this just looks
1444 		 * like a standard reply.
1445 		 */
1446 		complete(&knotif->ready);
1447 	}
1448 
1449 	seccomp_notify_free(filter);
1450 	mutex_unlock(&filter->notify_lock);
1451 }
1452 
1453 static int seccomp_notify_release(struct inode *inode, struct file *file)
1454 {
1455 	struct seccomp_filter *filter = file->private_data;
1456 
1457 	seccomp_notify_detach(filter);
1458 	__put_seccomp_filter(filter);
1459 	return 0;
1460 }
1461 
1462 /* must be called with notif_lock held */
1463 static inline struct seccomp_knotif *
1464 find_notification(struct seccomp_filter *filter, u64 id)
1465 {
1466 	struct seccomp_knotif *cur;
1467 
1468 	lockdep_assert_held(&filter->notify_lock);
1469 
1470 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1471 		if (cur->id == id)
1472 			return cur;
1473 	}
1474 
1475 	return NULL;
1476 }
1477 
1478 static int recv_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
1479 				  void *key)
1480 {
1481 	/* Avoid a wakeup if event not interesting for us. */
1482 	if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR | EPOLLHUP)))
1483 		return 0;
1484 	return autoremove_wake_function(wait, mode, sync, key);
1485 }
1486 
1487 static int recv_wait_event(struct seccomp_filter *filter)
1488 {
1489 	DEFINE_WAIT_FUNC(wait, recv_wake_function);
1490 	int ret;
1491 
1492 	if (refcount_read(&filter->users) == 0)
1493 		return 0;
1494 
1495 	if (atomic_dec_if_positive(&filter->notif->requests) >= 0)
1496 		return 0;
1497 
1498 	for (;;) {
1499 		ret = prepare_to_wait_event(&filter->wqh, &wait, TASK_INTERRUPTIBLE);
1500 
1501 		if (atomic_dec_if_positive(&filter->notif->requests) >= 0)
1502 			break;
1503 		if (refcount_read(&filter->users) == 0)
1504 			break;
1505 
1506 		if (ret)
1507 			return ret;
1508 
1509 		schedule();
1510 	}
1511 	finish_wait(&filter->wqh, &wait);
1512 	return 0;
1513 }
1514 
1515 static long seccomp_notify_recv(struct seccomp_filter *filter,
1516 				void __user *buf)
1517 {
1518 	struct seccomp_knotif *knotif = NULL, *cur;
1519 	struct seccomp_notif unotif;
1520 	ssize_t ret;
1521 
1522 	/* Verify that we're not given garbage to keep struct extensible. */
1523 	ret = check_zeroed_user(buf, sizeof(unotif));
1524 	if (ret < 0)
1525 		return ret;
1526 	if (!ret)
1527 		return -EINVAL;
1528 
1529 	memset(&unotif, 0, sizeof(unotif));
1530 
1531 	ret = recv_wait_event(filter);
1532 	if (ret < 0)
1533 		return ret;
1534 
1535 	mutex_lock(&filter->notify_lock);
1536 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1537 		if (cur->state == SECCOMP_NOTIFY_INIT) {
1538 			knotif = cur;
1539 			break;
1540 		}
1541 	}
1542 
1543 	/*
1544 	 * If we didn't find a notification, it could be that the task was
1545 	 * interrupted by a fatal signal between the time we were woken and
1546 	 * when we were able to acquire the rw lock.
1547 	 */
1548 	if (!knotif) {
1549 		ret = -ENOENT;
1550 		goto out;
1551 	}
1552 
1553 	unotif.id = knotif->id;
1554 	unotif.pid = task_pid_vnr(knotif->task);
1555 	unotif.data = *(knotif->data);
1556 
1557 	knotif->state = SECCOMP_NOTIFY_SENT;
1558 	wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1559 	ret = 0;
1560 out:
1561 	mutex_unlock(&filter->notify_lock);
1562 
1563 	if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1564 		ret = -EFAULT;
1565 
1566 		/*
1567 		 * Userspace screwed up. To make sure that we keep this
1568 		 * notification alive, let's reset it back to INIT. It
1569 		 * may have died when we released the lock, so we need to make
1570 		 * sure it's still around.
1571 		 */
1572 		mutex_lock(&filter->notify_lock);
1573 		knotif = find_notification(filter, unotif.id);
1574 		if (knotif) {
1575 			/* Reset the process to make sure it's not stuck */
1576 			if (should_sleep_killable(filter, knotif))
1577 				complete(&knotif->ready);
1578 			knotif->state = SECCOMP_NOTIFY_INIT;
1579 			atomic_inc(&filter->notif->requests);
1580 			wake_up_poll(&filter->wqh, EPOLLIN | EPOLLRDNORM);
1581 		}
1582 		mutex_unlock(&filter->notify_lock);
1583 	}
1584 
1585 	return ret;
1586 }
1587 
1588 static long seccomp_notify_send(struct seccomp_filter *filter,
1589 				void __user *buf)
1590 {
1591 	struct seccomp_notif_resp resp = {};
1592 	struct seccomp_knotif *knotif;
1593 	long ret;
1594 
1595 	if (copy_from_user(&resp, buf, sizeof(resp)))
1596 		return -EFAULT;
1597 
1598 	if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1599 		return -EINVAL;
1600 
1601 	if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1602 	    (resp.error || resp.val))
1603 		return -EINVAL;
1604 
1605 	ret = mutex_lock_interruptible(&filter->notify_lock);
1606 	if (ret < 0)
1607 		return ret;
1608 
1609 	knotif = find_notification(filter, resp.id);
1610 	if (!knotif) {
1611 		ret = -ENOENT;
1612 		goto out;
1613 	}
1614 
1615 	/* Allow exactly one reply. */
1616 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1617 		ret = -EINPROGRESS;
1618 		goto out;
1619 	}
1620 
1621 	ret = 0;
1622 	knotif->state = SECCOMP_NOTIFY_REPLIED;
1623 	knotif->error = resp.error;
1624 	knotif->val = resp.val;
1625 	knotif->flags = resp.flags;
1626 	if (filter->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1627 		complete_on_current_cpu(&knotif->ready);
1628 	else
1629 		complete(&knotif->ready);
1630 out:
1631 	mutex_unlock(&filter->notify_lock);
1632 	return ret;
1633 }
1634 
1635 static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1636 				    void __user *buf)
1637 {
1638 	struct seccomp_knotif *knotif;
1639 	u64 id;
1640 	long ret;
1641 
1642 	if (copy_from_user(&id, buf, sizeof(id)))
1643 		return -EFAULT;
1644 
1645 	ret = mutex_lock_interruptible(&filter->notify_lock);
1646 	if (ret < 0)
1647 		return ret;
1648 
1649 	knotif = find_notification(filter, id);
1650 	if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1651 		ret = 0;
1652 	else
1653 		ret = -ENOENT;
1654 
1655 	mutex_unlock(&filter->notify_lock);
1656 	return ret;
1657 }
1658 
1659 static long seccomp_notify_set_flags(struct seccomp_filter *filter,
1660 				    unsigned long flags)
1661 {
1662 	long ret;
1663 
1664 	if (flags & ~SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP)
1665 		return -EINVAL;
1666 
1667 	ret = mutex_lock_interruptible(&filter->notify_lock);
1668 	if (ret < 0)
1669 		return ret;
1670 	filter->notif->flags = flags;
1671 	mutex_unlock(&filter->notify_lock);
1672 	return 0;
1673 }
1674 
1675 static long seccomp_notify_addfd(struct seccomp_filter *filter,
1676 				 struct seccomp_notif_addfd __user *uaddfd,
1677 				 unsigned int size)
1678 {
1679 	struct seccomp_notif_addfd addfd;
1680 	struct seccomp_knotif *knotif;
1681 	struct seccomp_kaddfd kaddfd;
1682 	int ret;
1683 
1684 	BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1685 	BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1686 
1687 	if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1688 		return -EINVAL;
1689 
1690 	ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1691 	if (ret)
1692 		return ret;
1693 
1694 	if (addfd.newfd_flags & ~O_CLOEXEC)
1695 		return -EINVAL;
1696 
1697 	if (addfd.flags & ~(SECCOMP_ADDFD_FLAG_SETFD | SECCOMP_ADDFD_FLAG_SEND))
1698 		return -EINVAL;
1699 
1700 	if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1701 		return -EINVAL;
1702 
1703 	kaddfd.file = fget(addfd.srcfd);
1704 	if (!kaddfd.file)
1705 		return -EBADF;
1706 
1707 	kaddfd.ioctl_flags = addfd.flags;
1708 	kaddfd.flags = addfd.newfd_flags;
1709 	kaddfd.setfd = addfd.flags & SECCOMP_ADDFD_FLAG_SETFD;
1710 	kaddfd.fd = addfd.newfd;
1711 	init_completion(&kaddfd.completion);
1712 
1713 	ret = mutex_lock_interruptible(&filter->notify_lock);
1714 	if (ret < 0)
1715 		goto out;
1716 
1717 	knotif = find_notification(filter, addfd.id);
1718 	if (!knotif) {
1719 		ret = -ENOENT;
1720 		goto out_unlock;
1721 	}
1722 
1723 	/*
1724 	 * We do not want to allow for FD injection to occur before the
1725 	 * notification has been picked up by a userspace handler, or after
1726 	 * the notification has been replied to.
1727 	 */
1728 	if (knotif->state != SECCOMP_NOTIFY_SENT) {
1729 		ret = -EINPROGRESS;
1730 		goto out_unlock;
1731 	}
1732 
1733 	if (addfd.flags & SECCOMP_ADDFD_FLAG_SEND) {
1734 		/*
1735 		 * Disallow queuing an atomic addfd + send reply while there are
1736 		 * some addfd requests still to process.
1737 		 *
1738 		 * There is no clear reason to support it and allows us to keep
1739 		 * the loop on the other side straight-forward.
1740 		 */
1741 		if (!list_empty(&knotif->addfd)) {
1742 			ret = -EBUSY;
1743 			goto out_unlock;
1744 		}
1745 
1746 		/* Allow exactly only one reply */
1747 		knotif->state = SECCOMP_NOTIFY_REPLIED;
1748 	}
1749 
1750 	list_add(&kaddfd.list, &knotif->addfd);
1751 	complete(&knotif->ready);
1752 	mutex_unlock(&filter->notify_lock);
1753 
1754 	/* Now we wait for it to be processed or be interrupted */
1755 	ret = wait_for_completion_interruptible(&kaddfd.completion);
1756 	if (ret == 0) {
1757 		/*
1758 		 * We had a successful completion. The other side has already
1759 		 * removed us from the addfd queue, and
1760 		 * wait_for_completion_interruptible has a memory barrier upon
1761 		 * success that lets us read this value directly without
1762 		 * locking.
1763 		 */
1764 		ret = kaddfd.ret;
1765 		goto out;
1766 	}
1767 
1768 	mutex_lock(&filter->notify_lock);
1769 	/*
1770 	 * Even though we were woken up by a signal and not a successful
1771 	 * completion, a completion may have happened in the mean time.
1772 	 *
1773 	 * We need to check again if the addfd request has been handled,
1774 	 * and if not, we will remove it from the queue.
1775 	 */
1776 	if (list_empty(&kaddfd.list))
1777 		ret = kaddfd.ret;
1778 	else
1779 		list_del(&kaddfd.list);
1780 
1781 out_unlock:
1782 	mutex_unlock(&filter->notify_lock);
1783 out:
1784 	fput(kaddfd.file);
1785 
1786 	return ret;
1787 }
1788 
1789 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1790 				 unsigned long arg)
1791 {
1792 	struct seccomp_filter *filter = file->private_data;
1793 	void __user *buf = (void __user *)arg;
1794 
1795 	/* Fixed-size ioctls */
1796 	switch (cmd) {
1797 	case SECCOMP_IOCTL_NOTIF_RECV:
1798 		return seccomp_notify_recv(filter, buf);
1799 	case SECCOMP_IOCTL_NOTIF_SEND:
1800 		return seccomp_notify_send(filter, buf);
1801 	case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1802 	case SECCOMP_IOCTL_NOTIF_ID_VALID:
1803 		return seccomp_notify_id_valid(filter, buf);
1804 	case SECCOMP_IOCTL_NOTIF_SET_FLAGS:
1805 		return seccomp_notify_set_flags(filter, arg);
1806 	}
1807 
1808 	/* Extensible Argument ioctls */
1809 #define EA_IOCTL(cmd)	((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1810 	switch (EA_IOCTL(cmd)) {
1811 	case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1812 		return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1813 	default:
1814 		return -EINVAL;
1815 	}
1816 }
1817 
1818 static __poll_t seccomp_notify_poll(struct file *file,
1819 				    struct poll_table_struct *poll_tab)
1820 {
1821 	struct seccomp_filter *filter = file->private_data;
1822 	__poll_t ret = 0;
1823 	struct seccomp_knotif *cur;
1824 
1825 	poll_wait(file, &filter->wqh, poll_tab);
1826 
1827 	if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1828 		return EPOLLERR;
1829 
1830 	list_for_each_entry(cur, &filter->notif->notifications, list) {
1831 		if (cur->state == SECCOMP_NOTIFY_INIT)
1832 			ret |= EPOLLIN | EPOLLRDNORM;
1833 		if (cur->state == SECCOMP_NOTIFY_SENT)
1834 			ret |= EPOLLOUT | EPOLLWRNORM;
1835 		if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1836 			break;
1837 	}
1838 
1839 	mutex_unlock(&filter->notify_lock);
1840 
1841 	if (refcount_read(&filter->users) == 0)
1842 		ret |= EPOLLHUP;
1843 
1844 	return ret;
1845 }
1846 
1847 static const struct file_operations seccomp_notify_ops = {
1848 	.poll = seccomp_notify_poll,
1849 	.release = seccomp_notify_release,
1850 	.unlocked_ioctl = seccomp_notify_ioctl,
1851 	.compat_ioctl = seccomp_notify_ioctl,
1852 };
1853 
1854 static struct file *init_listener(struct seccomp_filter *filter)
1855 {
1856 	struct file *ret;
1857 
1858 	ret = ERR_PTR(-ENOMEM);
1859 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1860 	if (!filter->notif)
1861 		goto out;
1862 
1863 	filter->notif->next_id = get_random_u64();
1864 	INIT_LIST_HEAD(&filter->notif->notifications);
1865 
1866 	ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1867 				 filter, O_RDWR);
1868 	if (IS_ERR(ret))
1869 		goto out_notif;
1870 
1871 	/* The file has a reference to it now */
1872 	__get_seccomp_filter(filter);
1873 
1874 out_notif:
1875 	if (IS_ERR(ret))
1876 		seccomp_notify_free(filter);
1877 out:
1878 	return ret;
1879 }
1880 
1881 /*
1882  * Does @new_child have a listener while an ancestor also has a listener?
1883  * If so, we'll want to reject this filter.
1884  * This only has to be tested for the current process, even in the TSYNC case,
1885  * because TSYNC installs @child with the same parent on all threads.
1886  * Note that @new_child is not hooked up to its parent at this point yet, so
1887  * we use current->seccomp.filter.
1888  */
1889 static bool has_duplicate_listener(struct seccomp_filter *new_child)
1890 {
1891 	struct seccomp_filter *cur;
1892 
1893 	/* must be protected against concurrent TSYNC */
1894 	lockdep_assert_held(&current->sighand->siglock);
1895 
1896 	if (!new_child->notif)
1897 		return false;
1898 	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1899 		if (cur->notif)
1900 			return true;
1901 	}
1902 
1903 	return false;
1904 }
1905 
1906 /**
1907  * seccomp_set_mode_filter: internal function for setting seccomp filter
1908  * @flags:  flags to change filter behavior
1909  * @filter: struct sock_fprog containing filter
1910  *
1911  * This function may be called repeatedly to install additional filters.
1912  * Every filter successfully installed will be evaluated (in reverse order)
1913  * for each system call the task makes.
1914  *
1915  * Once current->seccomp.mode is non-zero, it may not be changed.
1916  *
1917  * Returns 0 on success or -EINVAL on failure.
1918  */
1919 static long seccomp_set_mode_filter(unsigned int flags,
1920 				    const char __user *filter)
1921 {
1922 	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1923 	struct seccomp_filter *prepared = NULL;
1924 	long ret = -EINVAL;
1925 	int listener = -1;
1926 	struct file *listener_f = NULL;
1927 
1928 	/* Validate flags. */
1929 	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1930 		return -EINVAL;
1931 
1932 	/*
1933 	 * In the successful case, NEW_LISTENER returns the new listener fd.
1934 	 * But in the failure case, TSYNC returns the thread that died. If you
1935 	 * combine these two flags, there's no way to tell whether something
1936 	 * succeeded or failed. So, let's disallow this combination if the user
1937 	 * has not explicitly requested no errors from TSYNC.
1938 	 */
1939 	if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1940 	    (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1941 	    ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1942 		return -EINVAL;
1943 
1944 	/*
1945 	 * The SECCOMP_FILTER_FLAG_WAIT_KILLABLE_SENT flag doesn't make sense
1946 	 * without the SECCOMP_FILTER_FLAG_NEW_LISTENER flag.
1947 	 */
1948 	if ((flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV) &&
1949 	    ((flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) == 0))
1950 		return -EINVAL;
1951 
1952 	/* Prepare the new filter before holding any locks. */
1953 	prepared = seccomp_prepare_user_filter(filter);
1954 	if (IS_ERR(prepared))
1955 		return PTR_ERR(prepared);
1956 
1957 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1958 		listener = get_unused_fd_flags(O_CLOEXEC);
1959 		if (listener < 0) {
1960 			ret = listener;
1961 			goto out_free;
1962 		}
1963 
1964 		listener_f = init_listener(prepared);
1965 		if (IS_ERR(listener_f)) {
1966 			put_unused_fd(listener);
1967 			ret = PTR_ERR(listener_f);
1968 			goto out_free;
1969 		}
1970 	}
1971 
1972 	/*
1973 	 * Make sure we cannot change seccomp or nnp state via TSYNC
1974 	 * while another thread is in the middle of calling exec.
1975 	 */
1976 	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1977 	    mutex_lock_killable(&current->signal->cred_guard_mutex))
1978 		goto out_put_fd;
1979 
1980 	spin_lock_irq(&current->sighand->siglock);
1981 
1982 	if (!seccomp_may_assign_mode(seccomp_mode))
1983 		goto out;
1984 
1985 	if (has_duplicate_listener(prepared)) {
1986 		ret = -EBUSY;
1987 		goto out;
1988 	}
1989 
1990 	ret = seccomp_attach_filter(flags, prepared);
1991 	if (ret)
1992 		goto out;
1993 	/* Do not free the successfully attached filter. */
1994 	prepared = NULL;
1995 
1996 	seccomp_assign_mode(current, seccomp_mode, flags);
1997 out:
1998 	spin_unlock_irq(&current->sighand->siglock);
1999 	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
2000 		mutex_unlock(&current->signal->cred_guard_mutex);
2001 out_put_fd:
2002 	if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
2003 		if (ret) {
2004 			listener_f->private_data = NULL;
2005 			fput(listener_f);
2006 			put_unused_fd(listener);
2007 			seccomp_notify_detach(prepared);
2008 		} else {
2009 			fd_install(listener, listener_f);
2010 			ret = listener;
2011 		}
2012 	}
2013 out_free:
2014 	seccomp_filter_free(prepared);
2015 	return ret;
2016 }
2017 #else
2018 static inline long seccomp_set_mode_filter(unsigned int flags,
2019 					   const char __user *filter)
2020 {
2021 	return -EINVAL;
2022 }
2023 #endif
2024 
2025 static long seccomp_get_action_avail(const char __user *uaction)
2026 {
2027 	u32 action;
2028 
2029 	if (copy_from_user(&action, uaction, sizeof(action)))
2030 		return -EFAULT;
2031 
2032 	switch (action) {
2033 	case SECCOMP_RET_KILL_PROCESS:
2034 	case SECCOMP_RET_KILL_THREAD:
2035 	case SECCOMP_RET_TRAP:
2036 	case SECCOMP_RET_ERRNO:
2037 	case SECCOMP_RET_USER_NOTIF:
2038 	case SECCOMP_RET_TRACE:
2039 	case SECCOMP_RET_LOG:
2040 	case SECCOMP_RET_ALLOW:
2041 		break;
2042 	default:
2043 		return -EOPNOTSUPP;
2044 	}
2045 
2046 	return 0;
2047 }
2048 
2049 static long seccomp_get_notif_sizes(void __user *usizes)
2050 {
2051 	struct seccomp_notif_sizes sizes = {
2052 		.seccomp_notif = sizeof(struct seccomp_notif),
2053 		.seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
2054 		.seccomp_data = sizeof(struct seccomp_data),
2055 	};
2056 
2057 	if (copy_to_user(usizes, &sizes, sizeof(sizes)))
2058 		return -EFAULT;
2059 
2060 	return 0;
2061 }
2062 
2063 /* Common entry point for both prctl and syscall. */
2064 static long do_seccomp(unsigned int op, unsigned int flags,
2065 		       void __user *uargs)
2066 {
2067 	switch (op) {
2068 	case SECCOMP_SET_MODE_STRICT:
2069 		if (flags != 0 || uargs != NULL)
2070 			return -EINVAL;
2071 		return seccomp_set_mode_strict();
2072 	case SECCOMP_SET_MODE_FILTER:
2073 		return seccomp_set_mode_filter(flags, uargs);
2074 	case SECCOMP_GET_ACTION_AVAIL:
2075 		if (flags != 0)
2076 			return -EINVAL;
2077 
2078 		return seccomp_get_action_avail(uargs);
2079 	case SECCOMP_GET_NOTIF_SIZES:
2080 		if (flags != 0)
2081 			return -EINVAL;
2082 
2083 		return seccomp_get_notif_sizes(uargs);
2084 	default:
2085 		return -EINVAL;
2086 	}
2087 }
2088 
2089 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
2090 			 void __user *, uargs)
2091 {
2092 	return do_seccomp(op, flags, uargs);
2093 }
2094 
2095 /**
2096  * prctl_set_seccomp: configures current->seccomp.mode
2097  * @seccomp_mode: requested mode to use
2098  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
2099  *
2100  * Returns 0 on success or -EINVAL on failure.
2101  */
2102 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
2103 {
2104 	unsigned int op;
2105 	void __user *uargs;
2106 
2107 	switch (seccomp_mode) {
2108 	case SECCOMP_MODE_STRICT:
2109 		op = SECCOMP_SET_MODE_STRICT;
2110 		/*
2111 		 * Setting strict mode through prctl always ignored filter,
2112 		 * so make sure it is always NULL here to pass the internal
2113 		 * check in do_seccomp().
2114 		 */
2115 		uargs = NULL;
2116 		break;
2117 	case SECCOMP_MODE_FILTER:
2118 		op = SECCOMP_SET_MODE_FILTER;
2119 		uargs = filter;
2120 		break;
2121 	default:
2122 		return -EINVAL;
2123 	}
2124 
2125 	/* prctl interface doesn't have flags, so they are always zero. */
2126 	return do_seccomp(op, 0, uargs);
2127 }
2128 
2129 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
2130 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
2131 					     unsigned long filter_off)
2132 {
2133 	struct seccomp_filter *orig, *filter;
2134 	unsigned long count;
2135 
2136 	/*
2137 	 * Note: this is only correct because the caller should be the (ptrace)
2138 	 * tracer of the task, otherwise lock_task_sighand is needed.
2139 	 */
2140 	spin_lock_irq(&task->sighand->siglock);
2141 
2142 	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
2143 		spin_unlock_irq(&task->sighand->siglock);
2144 		return ERR_PTR(-EINVAL);
2145 	}
2146 
2147 	orig = task->seccomp.filter;
2148 	__get_seccomp_filter(orig);
2149 	spin_unlock_irq(&task->sighand->siglock);
2150 
2151 	count = 0;
2152 	for (filter = orig; filter; filter = filter->prev)
2153 		count++;
2154 
2155 	if (filter_off >= count) {
2156 		filter = ERR_PTR(-ENOENT);
2157 		goto out;
2158 	}
2159 
2160 	count -= filter_off;
2161 	for (filter = orig; filter && count > 1; filter = filter->prev)
2162 		count--;
2163 
2164 	if (WARN_ON(count != 1 || !filter)) {
2165 		filter = ERR_PTR(-ENOENT);
2166 		goto out;
2167 	}
2168 
2169 	__get_seccomp_filter(filter);
2170 
2171 out:
2172 	__put_seccomp_filter(orig);
2173 	return filter;
2174 }
2175 
2176 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
2177 			void __user *data)
2178 {
2179 	struct seccomp_filter *filter;
2180 	struct sock_fprog_kern *fprog;
2181 	long ret;
2182 
2183 	if (!capable(CAP_SYS_ADMIN) ||
2184 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2185 		return -EACCES;
2186 	}
2187 
2188 	filter = get_nth_filter(task, filter_off);
2189 	if (IS_ERR(filter))
2190 		return PTR_ERR(filter);
2191 
2192 	fprog = filter->prog->orig_prog;
2193 	if (!fprog) {
2194 		/* This must be a new non-cBPF filter, since we save
2195 		 * every cBPF filter's orig_prog above when
2196 		 * CONFIG_CHECKPOINT_RESTORE is enabled.
2197 		 */
2198 		ret = -EMEDIUMTYPE;
2199 		goto out;
2200 	}
2201 
2202 	ret = fprog->len;
2203 	if (!data)
2204 		goto out;
2205 
2206 	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
2207 		ret = -EFAULT;
2208 
2209 out:
2210 	__put_seccomp_filter(filter);
2211 	return ret;
2212 }
2213 
2214 long seccomp_get_metadata(struct task_struct *task,
2215 			  unsigned long size, void __user *data)
2216 {
2217 	long ret;
2218 	struct seccomp_filter *filter;
2219 	struct seccomp_metadata kmd = {};
2220 
2221 	if (!capable(CAP_SYS_ADMIN) ||
2222 	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
2223 		return -EACCES;
2224 	}
2225 
2226 	size = min_t(unsigned long, size, sizeof(kmd));
2227 
2228 	if (size < sizeof(kmd.filter_off))
2229 		return -EINVAL;
2230 
2231 	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
2232 		return -EFAULT;
2233 
2234 	filter = get_nth_filter(task, kmd.filter_off);
2235 	if (IS_ERR(filter))
2236 		return PTR_ERR(filter);
2237 
2238 	if (filter->log)
2239 		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
2240 
2241 	ret = size;
2242 	if (copy_to_user(data, &kmd, size))
2243 		ret = -EFAULT;
2244 
2245 	__put_seccomp_filter(filter);
2246 	return ret;
2247 }
2248 #endif
2249 
2250 #ifdef CONFIG_SYSCTL
2251 
2252 /* Human readable action names for friendly sysctl interaction */
2253 #define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
2254 #define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
2255 #define SECCOMP_RET_TRAP_NAME		"trap"
2256 #define SECCOMP_RET_ERRNO_NAME		"errno"
2257 #define SECCOMP_RET_USER_NOTIF_NAME	"user_notif"
2258 #define SECCOMP_RET_TRACE_NAME		"trace"
2259 #define SECCOMP_RET_LOG_NAME		"log"
2260 #define SECCOMP_RET_ALLOW_NAME		"allow"
2261 
2262 static const char seccomp_actions_avail[] =
2263 				SECCOMP_RET_KILL_PROCESS_NAME	" "
2264 				SECCOMP_RET_KILL_THREAD_NAME	" "
2265 				SECCOMP_RET_TRAP_NAME		" "
2266 				SECCOMP_RET_ERRNO_NAME		" "
2267 				SECCOMP_RET_USER_NOTIF_NAME     " "
2268 				SECCOMP_RET_TRACE_NAME		" "
2269 				SECCOMP_RET_LOG_NAME		" "
2270 				SECCOMP_RET_ALLOW_NAME;
2271 
2272 struct seccomp_log_name {
2273 	u32		log;
2274 	const char	*name;
2275 };
2276 
2277 static const struct seccomp_log_name seccomp_log_names[] = {
2278 	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
2279 	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
2280 	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
2281 	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
2282 	{ SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
2283 	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
2284 	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
2285 	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
2286 	{ }
2287 };
2288 
2289 static bool seccomp_names_from_actions_logged(char *names, size_t size,
2290 					      u32 actions_logged,
2291 					      const char *sep)
2292 {
2293 	const struct seccomp_log_name *cur;
2294 	bool append_sep = false;
2295 
2296 	for (cur = seccomp_log_names; cur->name && size; cur++) {
2297 		ssize_t ret;
2298 
2299 		if (!(actions_logged & cur->log))
2300 			continue;
2301 
2302 		if (append_sep) {
2303 			ret = strscpy(names, sep, size);
2304 			if (ret < 0)
2305 				return false;
2306 
2307 			names += ret;
2308 			size -= ret;
2309 		} else
2310 			append_sep = true;
2311 
2312 		ret = strscpy(names, cur->name, size);
2313 		if (ret < 0)
2314 			return false;
2315 
2316 		names += ret;
2317 		size -= ret;
2318 	}
2319 
2320 	return true;
2321 }
2322 
2323 static bool seccomp_action_logged_from_name(u32 *action_logged,
2324 					    const char *name)
2325 {
2326 	const struct seccomp_log_name *cur;
2327 
2328 	for (cur = seccomp_log_names; cur->name; cur++) {
2329 		if (!strcmp(cur->name, name)) {
2330 			*action_logged = cur->log;
2331 			return true;
2332 		}
2333 	}
2334 
2335 	return false;
2336 }
2337 
2338 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
2339 {
2340 	char *name;
2341 
2342 	*actions_logged = 0;
2343 	while ((name = strsep(&names, " ")) && *name) {
2344 		u32 action_logged = 0;
2345 
2346 		if (!seccomp_action_logged_from_name(&action_logged, name))
2347 			return false;
2348 
2349 		*actions_logged |= action_logged;
2350 	}
2351 
2352 	return true;
2353 }
2354 
2355 static int read_actions_logged(const struct ctl_table *ro_table, void *buffer,
2356 			       size_t *lenp, loff_t *ppos)
2357 {
2358 	char names[sizeof(seccomp_actions_avail)];
2359 	struct ctl_table table;
2360 
2361 	memset(names, 0, sizeof(names));
2362 
2363 	if (!seccomp_names_from_actions_logged(names, sizeof(names),
2364 					       seccomp_actions_logged, " "))
2365 		return -EINVAL;
2366 
2367 	table = *ro_table;
2368 	table.data = names;
2369 	table.maxlen = sizeof(names);
2370 	return proc_dostring(&table, 0, buffer, lenp, ppos);
2371 }
2372 
2373 static int write_actions_logged(const struct ctl_table *ro_table, void *buffer,
2374 				size_t *lenp, loff_t *ppos, u32 *actions_logged)
2375 {
2376 	char names[sizeof(seccomp_actions_avail)];
2377 	struct ctl_table table;
2378 	int ret;
2379 
2380 	if (!capable(CAP_SYS_ADMIN))
2381 		return -EPERM;
2382 
2383 	memset(names, 0, sizeof(names));
2384 
2385 	table = *ro_table;
2386 	table.data = names;
2387 	table.maxlen = sizeof(names);
2388 	ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2389 	if (ret)
2390 		return ret;
2391 
2392 	if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2393 		return -EINVAL;
2394 
2395 	if (*actions_logged & SECCOMP_LOG_ALLOW)
2396 		return -EINVAL;
2397 
2398 	seccomp_actions_logged = *actions_logged;
2399 	return 0;
2400 }
2401 
2402 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2403 				 int ret)
2404 {
2405 	char names[sizeof(seccomp_actions_avail)];
2406 	char old_names[sizeof(seccomp_actions_avail)];
2407 	const char *new = names;
2408 	const char *old = old_names;
2409 
2410 	if (!audit_enabled)
2411 		return;
2412 
2413 	memset(names, 0, sizeof(names));
2414 	memset(old_names, 0, sizeof(old_names));
2415 
2416 	if (ret)
2417 		new = "?";
2418 	else if (!actions_logged)
2419 		new = "(none)";
2420 	else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2421 						    actions_logged, ","))
2422 		new = "?";
2423 
2424 	if (!old_actions_logged)
2425 		old = "(none)";
2426 	else if (!seccomp_names_from_actions_logged(old_names,
2427 						    sizeof(old_names),
2428 						    old_actions_logged, ","))
2429 		old = "?";
2430 
2431 	return audit_seccomp_actions_logged(new, old, !ret);
2432 }
2433 
2434 static int seccomp_actions_logged_handler(const struct ctl_table *ro_table, int write,
2435 					  void *buffer, size_t *lenp,
2436 					  loff_t *ppos)
2437 {
2438 	int ret;
2439 
2440 	if (write) {
2441 		u32 actions_logged = 0;
2442 		u32 old_actions_logged = seccomp_actions_logged;
2443 
2444 		ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2445 					   &actions_logged);
2446 		audit_actions_logged(actions_logged, old_actions_logged, ret);
2447 	} else
2448 		ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2449 
2450 	return ret;
2451 }
2452 
2453 static struct ctl_table seccomp_sysctl_table[] = {
2454 	{
2455 		.procname	= "actions_avail",
2456 		.data		= (void *) &seccomp_actions_avail,
2457 		.maxlen		= sizeof(seccomp_actions_avail),
2458 		.mode		= 0444,
2459 		.proc_handler	= proc_dostring,
2460 	},
2461 	{
2462 		.procname	= "actions_logged",
2463 		.mode		= 0644,
2464 		.proc_handler	= seccomp_actions_logged_handler,
2465 	},
2466 };
2467 
2468 static int __init seccomp_sysctl_init(void)
2469 {
2470 	register_sysctl_init("kernel/seccomp", seccomp_sysctl_table);
2471 	return 0;
2472 }
2473 
2474 device_initcall(seccomp_sysctl_init)
2475 
2476 #endif /* CONFIG_SYSCTL */
2477 
2478 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
2479 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */
2480 static void proc_pid_seccomp_cache_arch(struct seq_file *m, const char *name,
2481 					const void *bitmap, size_t bitmap_size)
2482 {
2483 	int nr;
2484 
2485 	for (nr = 0; nr < bitmap_size; nr++) {
2486 		bool cached = test_bit(nr, bitmap);
2487 		char *status = cached ? "ALLOW" : "FILTER";
2488 
2489 		seq_printf(m, "%s %d %s\n", name, nr, status);
2490 	}
2491 }
2492 
2493 int proc_pid_seccomp_cache(struct seq_file *m, struct pid_namespace *ns,
2494 			   struct pid *pid, struct task_struct *task)
2495 {
2496 	struct seccomp_filter *f;
2497 	unsigned long flags;
2498 
2499 	/*
2500 	 * We don't want some sandboxed process to know what their seccomp
2501 	 * filters consist of.
2502 	 */
2503 	if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
2504 		return -EACCES;
2505 
2506 	if (!lock_task_sighand(task, &flags))
2507 		return -ESRCH;
2508 
2509 	f = READ_ONCE(task->seccomp.filter);
2510 	if (!f) {
2511 		unlock_task_sighand(task, &flags);
2512 		return 0;
2513 	}
2514 
2515 	/* prevent filter from being freed while we are printing it */
2516 	__get_seccomp_filter(f);
2517 	unlock_task_sighand(task, &flags);
2518 
2519 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_NATIVE_NAME,
2520 				    f->cache.allow_native,
2521 				    SECCOMP_ARCH_NATIVE_NR);
2522 
2523 #ifdef SECCOMP_ARCH_COMPAT
2524 	proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_COMPAT_NAME,
2525 				    f->cache.allow_compat,
2526 				    SECCOMP_ARCH_COMPAT_NR);
2527 #endif /* SECCOMP_ARCH_COMPAT */
2528 
2529 	__put_seccomp_filter(f);
2530 	return 0;
2531 }
2532 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */
2533